Sim-to-Real Transfer for Quadrupedal Locomotion via Terrain Transformer

doi:10.1109/ICRA48891.2023.10160497

KMS > 创意与艺术学院

	Sim-to-Real Transfer for Quadrupedal Locomotion via Terrain Transformer
	Hang Lai 1; Weinan Zhang 1; Xialin He 1; Chen Yu 2; Zheng Tian3 ; Yong Yu 1; Jun Wang 2
	2023
会议录名称	PROCEEDINGS - IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION
ISSN	1050-4729
卷号	2023-May
页码	5141-5147
发表状态	已发表
DOI	10.1109/ICRA48891.2023.10160497
摘要	Deep reinforcement learning has recently emerged as an appealing alternative for legged locomotion over multiple terrains by training a policy in physical simulation and then transferring it to the real world (i.e., sim-to-real transfer). Despite considerable progress, the capacity and scalability of traditional neural networks are still limited, which may hinder their applications in more complex environments. In contrast, the Transformer architecture has shown its superiority in a wide range of large-scale sequence modeling tasks, including natural language processing and decision-making problems. In this paper, we propose Terrain Transformer (TERT), a high-capacity Transformer model for quadrupedal locomotion control on various terrains. Furthermore, to better leverage Transformer in sim-to-real scenarios, we present a novel two-stage training framework consisting of an offline pretraining stage and an online correction stage, which can naturally integrate Transformer with privileged training. Extensive experiments in simulation demonstrate that TERT outperforms state-of-the-art baselines on different terrains in terms of return, energy consumption and control smoothness. In further real-world validation, TERT successfully traverses nine challenging terrains, including sand pit and stair down, which can not be accomplished by strong baselines. © 2023 IEEE.
关键词	Deep learning Energy utilization Intelligent robots Modeling languages Natural language processing systems Reinforcement learning Complex environments Large-scale sequences Legged locomotion Modeling task Natural languages Neural-networks Physical simulation Real-world Reinforcement learnings Sequence models
会议名称	2023 IEEE International Conference on Robotics and Automation, ICRA 2023
会议地点	London, United kingdom
会议日期	May 29, 2023 - June 2, 2023
URL	查看原文
收录类别	EI
语种	英语
出版者	Institute of Electrical and Electronics Engineers Inc.
EI入藏号	20233514632357
EI主题词	Decision making
EI分类号	461.4 Ergonomics and Human Factors Engineering ; 525.3 Energy Utilization ; 723.2 Data Processing and Image Processing ; 723.4 Artificial Intelligence ; 731.6 Robot Applications ; 912.2 Management
原始文献类型	Conference article (CA)
来源库	IEEE
引用统计	正在获取...
文献类型	会议论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/325819
专题	创意与艺术学院创意与艺术学院_PI研究组(P)_田政组
作者单位	1.Dept. of Computer Sci. and Eng., Shanghai Jiao Tong University, China 2.Digital Brain Lab, Shanghai, China 3.School of Creativity and Art, ShanghaiTech University, China
推荐引用方式 GB/T 7714	Hang Lai,Weinan Zhang,Xialin He,et al. Sim-to-Real Transfer for Quadrupedal Locomotion via Terrain Transformer[C]:Institute of Electrical and Electronics Engineers Inc.,2023:5141-5147.